Direct Adaptive Neural Controller for Earthquake Excited Nonlinear Base Isolated Buildings

Abstract

Majority of the structural control schemes using ANN proposed so far in the literature involve a plant identifier, and a controller coupled with the identifier. Constructing a plant identifier requires significant training data, which is cumbersome for seismic structural applications. Direct adaptive control based on input-output representations is very attractive for such problems as the process of system identification is bypassed in the control formulation. A direct adaptive control scheme using neural networks is presented for the active control of earthquake excited nonlinear base isolated buildings. The control scheme is based on model reference direct adaptive control where the system response is made to follow a desired response. The active control force is approximated using a recently developed Extended Minimal Resource Allocation Network (EMRAN). Performance of the proposed control scheme is evaluated on a nonlinear three-dimensional base isolated benchmark structure, incorporating lateral-torsion superstructure behavior and the biaxial interaction of the nonlinear bearings at the isolation layer. The benchmark structure is highly complex due to strong coupling between the isolation level forces and the superstructure responses. Results are presented using a comprehensive set of the performance indices to realistically quantify the trade-offs associated with the control of nonlinear base isolated buildings. The results clearly show that EMRAN controller reduces the vibration level significantly for a wide range of earthquake records.